Design and Implement Product Metrics Dashboards

This skill is expert in designing and implementing product metrics dashboards that drive actionable insights, covering the full lifecycle from metric selection to visualization design. Its metric hierarchy distinguishes North Star metrics (primary business outcomes like MAU or revenue), leading indicators (predictive of future performance), and lagging indicators (confirm historical trends), always with cohort/time/product-area segmentation. It structures dashboards in three tiers - an executive summary (North Star metric and trend, 3-5 max KPIs, health check status), operational metrics (acquisition, engagement, retention, revenue), and diagnostic deep-dives (funnel analysis, cohort analysis, feature usage, user behavior). It provides SQL for weekly active user growth rate with week-over-week comparison and N-day retention cohort analysis (M0/M1/M3 user counts and retention percentages), a chart-type mapping matching metric type to visualization (line charts for trends, bar charts for comparisons, pie/donut for parts-of-whole, scatter plots for correlations, funnel charts for conversion), a consistent color-coding CSS system (green for positive, red for negative, amber for warning), a z-score-based anomaly detection function that alerts on metric spikes or drops outside a rolling-window expected range, an interactive filtering/drill-down controller, and a Redis-backed metrics cache with TTL-based invalidation.

Use this skill when building a product metrics dashboard that needs a real metric hierarchy, automated anomaly alerting, and performance-conscious data loading - not a flat collection of charts with no structure. Its dashboard governance guidance calls for clear metric ownership, refresh cadences matched to metric importance, role-based access control for sensitive metrics, and documented metric definitions and calculation methods. Its UX guidelines favor progressive disclosure (high-level first, drill-down for detail), mobile responsiveness, meaningful loading states, and graceful degradation on data unavailability. It is not meant for a one-off report - the skill defines dashboard success itself as a measured metric (adoption rate, time to insight, decision impact, data accuracy against source-of-truth validation), meaning the dashboard is expected to be an ongoing, governed product.

### Example: Automated anomaly detection
import pandas as pd
from scipy import stats

def detect_metric_anomalies(df, metric_column, threshold=2.0):
    """
    Detect anomalies in metrics using z-score analysis
    """
    df['rolling_mean'] = df[metric_column].rolling(window=7).mean()
    df['rolling_std'] = df[metric_column].rolling(window=7).std()
    df['z_score'] = (df[metric_column] - df['rolling_mean']) / df['rolling_std']
    
    anomalies = df[abs(df['z_score']) > threshold]
    
    for _, row in anomalies.iterrows():
        alert_type = 'spike' if row['z_score'] > 0 else 'drop'
        send_alert({
            'metric': metric_column,
            'date': row['date'],
            'value': row[metric_column],
            'type': alert_type
        })

Given a product's metric set, the skill produces a three-tier dashboard structure, acquisition and retention SQL like the growth-rate and cohort queries, a chart-type mapping and color-coding system for consistent visualization, the z-score anomaly detector shown above wired to an alerting function, a filtering/drill-down controller class, and a Redis-backed caching layer for metric queries.

Product analytics and data teams building dashboards that need to go beyond static charts into a governed, alerting, performant system - clear metric ownership, automated anomaly detection, and cached data loading. It suits teams that measure the dashboard itself (adoption, time to insight, decision impact, data accuracy) and want progressive-disclosure UX with mobile support and graceful error handling built in from the start.